Feature-based image stitching with repetitive patterns using local phase quantization

Stitching images with repeated patterns is one of the hard tasks in computer vision. Unfortunately, available algorithms do not have enough accuracy to register these kinds of images. This problem is due to lack of discriminative information of corresponding points of images. To overcome this shortcoming, a novel method for feature-based image stitching is proposed. In the proposed method, the SIFT descriptor is utilized to extract and match robust keypoints from each image. Texture information is achieved by using the local phase quantization (LPQ) around each keypoint. Then, a distinctive feature histogram is obtained based on LPQ. Finally, the symmetric Kullback-Leibler divergence (SKLD) is used as a dissimilarity measure between corresponding points to reject mismatches. Experimental results demonstrate that the proposed method can efficiently eliminate ambiguities in image stitching and successfully register images with repeated patterns.